Setup Guides
Practical setup guides for Agent Tracer. If you want to understand how the runtime and API contracts instead of how to install it, use the reference pages below.
Quick start
Agent Tracer currently has two guided setup paths:
- Claude Code plugin setup
- Codex bootstrap via repo-local hooks plus repo-local config
Minimal setup path
- Install and Run — clone the repo, install dependencies, start the monitor server and web dashboard, verify the installation.
- Claude Code Setup — load the plugin and register the MCP server. Claude Code integration is complete after these two steps.
- Codex Setup — use plain
codexwith the generated repo-local hooks and config.
Attach to external projects (optional)
If you want to use Agent Tracer with a project outside this repository, follow an additional step:
- External Project Setup — run
npm run setup:externalto generate.claude/settings.json,.codex/config.toml, and.codex/hooks.jsonin your target project.
Note: When running Claude Code inside the Agent Tracer repository itself,
setup:externalis not needed. You can start withclaude --plugin-dir packages/runtimedirectly.
Other runtimes (manual HTTP/MCP)
Runtimes other than Claude Code and Codex do not have automatic adapters. See External Project Setup § 6 for the minimal implementation order to call the HTTP API directly.
Reference
Once you're set up, these pages describe the event surface and runtime model in more detail.
- Runtime capabilities — how the capability registry tracks per-runtime differences
- API integration map — every HTTP endpoint mapped to hooks and manual-runtime use
- Runtime API flow & preprocessing — preprocessing applied inside each endpoint
- Claude Code hook payload spec — exact JSON shape of each hook payload consumed by the plugin
- Task observability — timeline-derived task summaries used by the dashboard
- SQLite schema — normalized SQLite table structure, event log, current/view tables, and migration rules
- Web styling guide — CSS / Tailwind conventions used by
@monitor/web
Related
- Runtime capabilities — runtime-specific behavior and evidence guarantees
- API integration map — HTTP and runtime endpoint coverage
- Task observability — diagnostics surfaced in the dashboard